Abstract
INTRODUCTION: Nursing services often face resource constraints, while reliable disease specific prioritization signals are difficult to obtain from limited routine data. We therefore developed PrepCare, a two-stage interpretable framework for infectious disease prioritization and alerting to support nursing resource planning. METHODS: PrepCare combines data-adaptive disease ranking with Seasonal-Trend decomposition using LOESS (STL)-based alerting. Diseases were prioritized using a multi-indicator composite index based on six pillars: burden, recent incidence, risk, trend, policy-informed severity, and a burden-severity interaction term. Pillar weights were assigned using an information entropy principle, and a Light Gradient Boosting Machine (LightGBM) regressor was used to capture non-linear interactions. SHapley Additive exPlanations for Tree-based models (TreeSHAP) were used to obtain a system-wide importance ranking. The top 15 diseases were then selected for alert assessment. STL was applied to log counts to remove seasonality and trend, and anomalies were detected from the residuals using calibrated thresholds. RESULTS: The framework produced a ranked list of priority diseases and generated alert signals for the top 15 diseases. STL-based alerting was supported by strong annual seasonal components, a reduction in residual variance, and decreased lag-1 autocorrelation after decomposition, indicating improved separation of recurrent patterns from irregular deviations. DISCUSSION: PrepCare offers a reproducible, interpretable, and label-agnostic "rank-selectalert" workflow for infectious disease early warning. The framework may support nursing staff scheduling, stock preparation, and risk communication, thereby enhancing preparedness in resource-constrained settings.